Journal of Memory and...

Journal of Memory and Language 84 (2015) 88–107

Contents lists available at ScienceDirect

Journal of Memory and Language

journal homepage: www.elsevier .com/locate / jml

Repetition reduction during word and concept overlapin bilingualsq

http://dx.doi.org/10.1016/j.jml.2015.05.0050749-596X/� 2015 Elsevier Inc. All rights reserved.

q The authors thank Ilsoo Hong, Euil Kim, Peter Kwak, Jaeryoung Lee,Ashlyn Sujin Park, and Si On Yoon for their assistance in constructingKorean stimuli and analyzing Korean speech data, and James Bartolotti,Susan Bobb, Sarah Chabal, Jen Krizman, Scott Schroeder, and AnthonyShook for comments on earlier drafts of the manuscript. This project wasfunded in part by training Grant NIDCD T32DC009399-04 to Tuan Lamand by Grant NICHD RO1HD059858 to Viorica Marian.⇑ Corresponding author at: Department of Communication Sciences

and Disorders, Northwestern University, 2240 Campus Drive, Evanston, IL60208, United States.

E-mail address: [email protected] (T.Q. Lam).

Tuan Q. Lam ⇑, Viorica MarianDepartment of Communication Sciences and Disorders, Northwestern University, United States

a r t i c l e i n f o a b s t r a c t

Article history:Received 26 January 2014revision received 16 May 2015

Keywords:ProminenceRepetitionSpeech productionLexical accessConceptBilingualism

In natural conversation, speakers often mention the same referents multiple times. Whilerepeated referents are produced with less prominence than non-repeated referents, it isunclear whether prominence reduction is due to repetition of concepts, words, or a com-bination of the two. In the current study, we dissociate these sources of repetition by exam-ining bilingual speakers, who have more than one word for the same concept across theirtwo languages. Three groups of Korean–English bilinguals (balanced, English-dominant,Korean-dominant) performed an event description task involving repetition of referentswithin a single language (i.e., repetition of word and concept) or across languages (i.e., rep-etition of concept only). While balanced bilinguals reduced prominence both within andacross languages, unbalanced bilinguals only reduced prominence when repetitionoccurred within a language. These patterns suggest that the degree to which words andconcepts are linked within a speaker’s language system determines the source of repetitionreduction.

� 2015 Elsevier Inc. All rights reserved.

Introduction

During conversation, people tend to mention a topicmultiple times, and this repetition affects the acoustic formof an utterance (Arnold, 1998). The first time a word occursin a conversation, production tends to be more exagger-ated. For example, you may say, ‘‘I have a cat” placingemphasis on the word ‘‘cat.” For the rest of the conversa-tion, you may be less careful with your pronunciation of

the word ‘‘cat,” and as a result, subsequent mentions of‘‘cat” will tend to have shorter duration and lower intensitycompared to the first production. This phenomenon isknown as repetition reduction. Repetition reduction is awell-characterized occurrence in which expressions thatrefer to repeated or given referents are produced with lessprominence (i.e., shorter duration and lower intensity)than expressions that refer to non-repeated or new refer-ents (Aylett & Turk, 2004; Bard et al., 2000; Bell, Brenier,Gregory, Girand, & Jurafsky, 2009; Fowler, 1988; Fowler& Housum, 1987; Galati & Brennan, 2010; Kahn & Arnold,2012; Lam & Watson, 2010; Pluymaekers, Ernestus, &Baayen, 2005; Watson, Arnold, & Tanenhaus, 2008).

While it is known that speakers reduce prominence forrepeated expressions, it is unknown where this repetitionreduction manifests along the production stream, becauserepetition can occur at multiple levels in the productionprocess (Baumann & Riester, 2013; Fowler, 1988; Lam &Watson, 2014). Though there are competing models of spo-ken word production, most models identify at least three

http://crossmark.crossref.org/dialog/?doi=10.1016/j.jml.2015.05.005&domain=pdf

http://dx.doi.org/10.1016/j.jml.2015.05.005

mailto:[email protected]


http://www.sciencedirect.com/science/journal/0749596X

http://www.elsevier.com/locate/jml

T.Q. Lam, V. Marian / Journal of Memory and Language 84 (2015) 88–107 89

important levels in the production process: a concept level,a word level, and a phoneme level1 (e.g., Costa, Caramazza,& Sebastian-Galles, 2000; Dell, 1986; Dell & O’Seaghdha,1991, 1992; Levelt, 1989; Levelt et al., 1999; for a review,see Rapp & Goldrick, 2000), and it is possible for repetitionreduction to take place at each of these levels. Fig. 1A–C pro-vide schematics of these levels of representation across dif-ferent models of production.

The production process begins with the activation ofconcepts representing a speaker’s intended meaning (e.g.,CAT). Activation from the concept level is then mappedonto corresponding words at the word level. At this point,multiple words may be active and the system needs tochoose which word(s) to send for further processing. Forexample, the concept of CAT may activate both the words‘‘cat” and ‘‘kitten” due to overlapping semantic informa-tion. Activation from the word level then flows downwardto activate phonemes for production (e.g., /k/, /æ/, /t/ for‘‘cat”). Within a normal conversation, repeatedly namingan object would involve repeated activation of the concept,repeated activation of the word, and repeated activation ofthe phonemes for a word. Theoretically, repetition reduc-tion could be rooted at any and all of these levels.

However, there is some evidence to suggest that repeti-tion reduction is not due to phoneme level repetition. In astudy by Fowler (1988), participants produced words thatcould be primed by either the same word or by a homo-phone. Homophones are words that completely overlapat the phoneme level (e.g., ‘‘seller” and ‘‘cellar”). Fowlerfound that while repeated words were reduced in promi-nence, words that were preceded by a homophone prime(i.e., phoneme repetition) were not reduced. This patternsuggests that repetition reduction is not due to repetitionat the phoneme level, but rather the result of repetitionat an earlier stage of the production process, specifically,at the levels of concepts or words.

Synonyms, or words that have nearly identicalmeaningsbut differ in form (e.g., couch and sofa), might offer a possi-ble approach to determining if repetition reduction is due torepetition at the concept level or repetition at the wordlevel. Indeed, synonyms were also tested for repetitionreduction in the Fowler (1988) study described above, butthe results were inconclusive because synonym primesresulted in durations that were in-between those of same-word primes (i.e., word repetition) and unrelated primes(i.e., no repetition), but that did not significantly differ fromeither prime type. This null resultmay be due to the fact thatFowler elicited synonym production via reading, as it is notclear that reading awordwould necessarily activate its syn-onyms in the same way as with more spontaneous produc-tion, such as object naming (e.g., Costa et al., 2000).Synonyms may also be problematic because speakers mayhave a context-dependent preference for one synonymoveranother, and having chosen to use a particular word, speak-ers have a tendency to reuse thatwordwhen referring to the

1 We will use the terms ‘‘concepts,” ‘‘words,” and ‘‘phonemes” through-out the paper, but these would correspond to ‘‘semantic features,” ‘‘lexicalnodes,” and ‘‘phonemes” in models by Dell and O’Seaghdha (1991) and Delland O’Seaghdha (1992) and to ‘‘lexical concepts,” ‘‘lemmas,” and ‘‘lexemes”in models by Levelt (1989) and Levelt, Roelofs, and Meyer (1999).

same referent (Brennan & Clark, 1996; Garrod & Anderson,1987). Moreover, the introduction of a new word may leadlisteners to interpret the word as referring to a differentobject (e.g., Metzing & Brennan, 2003) because, accordingto the principle of contrast, a difference in form would beexpected to mark a difference in meaning (Clark, 1990,1997). Even if listeners interpret the word as referring tothe same referent, a change of referring expressionmay leadlisteners to think of the referent differently, thereby alteringthe concept (Almor, 1999; Clark, 1990; Lam & Watson,2014). In both cases, the concept may have changed as aresult of the word change, making it difficult to separateconcept and word repetition.

Given the inconclusive findings from work with syn-onyms, it remains unclear whether repetition reductionis primarily driven by concept repetition, word repetition,or a combination of the two. Bilinguals, who are knownto code-switch between their two languages (Costa et al.,2000; Gollan & Ferreira, 2009; Kroll, Bobb, Misra, & Guo,2008), may offer another way to separate these two formsof repetition. Models of bilingual language production sug-gest that a bilingual’s two languages have mostly shared orhighly overlapping concepts but separate words for eachlanguage (e.g., Costa, 2005; Costa et al., 2000; Kroll &Stewart, 1994). Because bilinguals have more than oneword for any concept (i.e., one word for each language),it is possible to separate concept-level repetition fromword-level repetition. Unlike with synonyms, wherechanging the word may imply a different concept,changing from a word to its translation equivalent can bemotivated by a change in the response language whilepreserving the meaning. Indeed, Monsell, Matthews, andMiller (1992) used the language-switch paradigm to exam-ine whether prior production of a word primes productionfor its translation equivalent. In the first phase of theirstudy, word production was elicited from Welsh-Englishbilinguals via sentence completion (e.g., ‘‘a pine is a typeof ___”) in either Welsh (half the participants) or English(half the participants). Following the sentence completiontask, participants completed a separate picture-namingtask using Welsh. Speakers demonstrated repetition reduc-tion for words primed in the same language (Welsh?Welsh) but did not for words primed in a different lan-guage (English?Welsh). Given these results, Monsell,Matthews, and Miller argued that concept repetition aloneis not enough to elicit repetition priming.

While Monsell et al. (1992) did not find repetitionreduction across languages, their paradigm may not havebeen well suited for examining repetition reductionbecause the primes from the first phase of their study didnot describe the same referents as the words in the secondphase. In their study, primes were words that were elicitedvia definition prompts or via sentence completion, whichdiffer from the later picture-naming task. Thus, Monsell,Matthews, and Miller’s results cannot speak to whetherconceptually-driven repetition reduction will occur whenthere is also referent repetition. By contrast, in more natu-ral conversation, concept repetition usually occurs whenthe same referent is mentioned multiple times, meaningthat the method employed in this study involved less rep-etition of the referent than is typically observed in natural

Fig. 1. Schematics depicting the connections between the three major levels of representation for different models of word production (adapted from Rapp& Goldrick, 2000). The lines represent the connections between concepts and words and between words and phonemes. Fig. 1A presents the three majorlevels of production using terminology from Dell and O’Seaghdha (1991, 1992): semantic features map onto lexical nodes, which in turn map ontophonemes. Fig. 1B presents the three major levels using the terminology from Levelt (1989): lexical concepts map onto lemmas, which then map onto wordforms. Fig. 1C presents the three major levels using our unified terminology: concepts map onto words, which then map onto phonemes.

90 T.Q. Lam, V. Marian / Journal of Memory and Language 84 (2015) 88–107

speech. Additionally, their paradigm was suboptimalbecause it elicited isolated words rather than multi-wordphrases and sentences. Prior studies suggest that repetitionreduction effects are stronger in sentences (e.g., Fowler,1988), particularly when the word is in the same grammat-ical position (Terken & Hirschberg, 1994).

Finally, the lack of repetition reduction across languagesin Monsell et al. (1992) could be due to the language back-grounds of the participants in their study. Though it is pos-sible to separate concept-level repetition from word-levelrepetition in bilinguals, concepts may not be equally linkedto words across both languages. In the case of unbalancedbilinguals (that is, bilinguals with unequal proficiency intheir two languages), there may be stronger connectionsbetween concepts and words for the dominant languagethan for the less-dominant language, while in balancedbilinguals concepts may be equally linked to words in eachlanguage (Costa & Santesteban, 2004; Gollan & Ferreira,2009; Kroll & Stewart, 1994). Monsell et al. (1992) didnot report any information regarding language balancefor their bilinguals. If their bilinguals were unbalanced,prior activation from the concept level may not haveprimed words in both languages equally, which could haveled to the observed null effect of repetition in the languageswitch (concept repetition) condition. Furthermore, unbal-anced bilinguals may also have had difficulty switchingfrom using a word in the dominant language (English) tousing a word in a less dominant language (Welsh), whichmay have obscured any concept repetition effects. Weaddress these concerns by examining repetition reductionin both balanced and unbalanced bilinguals.

In our study, concept-level repetition and word-levelrepetition were dissociated by using a two-part eventdescription task in which the participant sequentiallyproduced two sentences that both referred to the sameobject. Repeated productions occurred both within (i.e.,both sentences are produced in the same language) andacross languages, (i.e., each sentence is produced in a differ-ent language). If concept-level repetition alone can lead to

repetition reduction, then speakers should reduce theprominence of repeated objects regardless of whether therepetition occurs across the bilingual’s two languages orwithin a single language. In contrast, if repetition reductionis due only to word-level repetition, then speakers shouldreduce the prominence of repeated objects only when theobject is repeatedwithin a single language. A third possibil-ity is that repetition operates at both the concept level andthe word level. In this case, there may be reduction forrepeated targets both when there is a language switch aswell as when repetition occurs within the same language,but the effect would be greater for repetition within thesame language.

As a secondary aim of the study, we examine whetherrepetition reduction is dependent on the connectionstrength between concepts and words by comparingbalanced vs. unbalanced bilinguals. Because languageimbalance may affect the relative level of activation forwords across the two languages, the degree to which con-cepts and words are linked within a language may impactrepetition reduction. While balanced bilinguals shouldhave relatively similar activation of words in bothlanguages for any given concept, unbalanced bilingualsmay preferentially activate words for the dominant languageover words in the less dominant language. Therefore, if rep-etition reduction is mediated by the availability of thewords, balanced bilinguals may show repetition reductionfor concept repetition by reducing prominence for repeatedobjects both within and across languages, while unbalancedbilinguals may only show an effect for word repetition,where repetition only leads to reduction within a singlelanguage, but not across languages.

Method

Participants

Forty-eight adult bilingual speakers of Korean andEnglish (mean age = 22.06; SD = 3.53) participated in the


experiment after providing informed consent. Participantshad lived in an English speaking country for an average of10.32 years (SD = 5.62) and lived in a Korean speakingcountry for an average of 11.73 years (SD = 6.71). All par-ticipants were university students currently studying andresiding in the United States. All participants completedthe Language Experience and Proficiency Questionnaire(LEAP-Q; Marian, Blumenfeld, & Kaushanskaya, 2007), aself-report questionnaire designed to assess the languageprofiles of multilingual participants. On a scale of 0–10,participants reported proficiencies (combined speakingand listening proficiency) of 9.05 (SD = 1.00) in Koreanand 8.96 (SD = 1.21) in English.

We separated participants into three groups: balancedbilinguals, Korean-dominant bilinguals, and English-dominant bilinguals. Language balance was determinedby relative proficiencies in Korean and English. Sixteenparticipants were classified as balanced bilinguals withsimilar proficiencies across Korean and English (a differ-ence of 1 or less; mean difference = 0.00, SD = 0.68, t(15)= 0.00, p > .10). Sixteen participants were classified asKorean-dominant bilinguals with Korean proficiency thatwas more than one point greater than their English profi-ciency (mean difference = 2.19; SD = 0.93, t(15) = 9.42,p < .001). The other sixteen participants were classified asEnglish-dominant bilinguals with English proficiency thatwas more than one point greater than their Korean profi-ciency (mean difference = 1.91; SD = 0.66, t(15) = 11.5,p < .001). There was a significant difference in proficiencybalance across the three groups (p’s < .001). Table 1 pre-sents the average proficiencies in Korean and English foreach of the three groups of bilinguals. Forty-four partici-pants listed Korean as the first language, one participantlisted English as the first language, and three participantslisted Korean and English as acquired simultaneously. Theaverage age of acquisition (AOA) for Korean was 0.50 years(SD = 0.77) and the average AOA for English was 5.71 years(SD = 3.66). Table 1 presents the average AOA of Koreanand English in each of the three groups. Individual partici-pant responses are presented in Appendix A.

Design

The study involved a two-part event description taskadapted from Lam and Watson (2010) in which partici-pants used two sentences to describe events involvingobjects displayed on a computer screen (e.g., ‘‘The appleis shrinking . . . the apple is flashing.”). The critical wordwas the name of the object in the second event. We manip-ulated three factors across each pair of sentences: (1)Whether or not the target object was the same in both

Table 1Participant characteristics.

Age of acquisition of Korean Age of acquis

Balanced 0.625 (0.72) 5.31 (3.86)Korean-Dominant 0.187 (0.40) 6.63 (4.35)English-Dominant 0.6875 (1.01) 5.19 (2.61)

Note: Average age of acquisition and average proficiency for balanced Koreanbilinguals as assessed using the Language Experience and Proficiency Questionnai

events; (2) Whether or not the target object in the secondevent was named in Korean or English; and (3) Whether ornot participants switched languages between the first andsecond events (henceforth: stay vs. code-switch trials).These three factors were crossed in a 2 (target object, samevs. different) � 2 (response-language, Korean vs. English) �2 (language-switch, stay vs. code-switch) design, leadingto eight conditions relating to the status of the seconditem.

Materials

This experiment used images from Rossion and Pourtois(2001), which are colored versions of the images fromSnodgrass and Vanderwart (1980). Of the 260 pictures inthe set, we removed images if their Korean names wereEnglish loan words or if their Korean names had initialphoneme overlap with their English counterparts (e.g., ‘‘펜” – [pen] and ‘‘pen”), resulting in 106 remaining images.Of the remaining images, 104 were used as targets in theexperiment: 96 in experimental trials and eight in practicetrials. From the 104 images used, we created 52 imagepairs such that the images’ names contained no initial pho-neme overlap within or between Korean and English. Eachimage pair served as the two possible targets for each trialand pairs were not repeated, resulting in a total of 52 trials.Of these 52 trials, four trials served as practice trials whilethe remaining 48 were critical trials. These 48 critical trialswere split into 24 Korean-response trials and 24 English-response trials. The mean syllable length of the 24 targetwords in Korean including the case marker was 3.23(0.67) syllables. In English, the mean syllable length forthe 24 target words was 1.73 (0.80). Conditions were pre-sented in a pseudorandomized order such that each of oureight conditions appeared exactly once in each set of eighttrials (i.e., Trials 1–8, Trials 9–16, etc.). As there were 48critical trials and eight conditions, participants saw eachcondition exactly six times over the course of the experi-ment. This set of 48 trials constituted the first list of itemsand conditions. We created another three lists of 48 trialsby counterbalancing repetition and code-switching withinitems across our trial lists. Another four lists were gener-ated by inverting the order of items and conditions in theprevious lists, leading to a total of eight different lists ofitems and conditions.

The language of response was elicited by either a pic-ture of a Korean actress (Son Yejin) or an American actress(Jennifer Aniston). Although no true addressee was pre-sent, participants were verbally instructed to treat theirresponse as being directed toward the person depicted.Hence, speakers were asked to use Korean when speaking

ition of English Korean proficiency English proficiency

9.41 (0.61) 9.41 (0.82)9.84 (0.30) 7.66 (0.96)7.91 (0.66) 9.81 (0.44)

–English bilinguals, Korean-dominant bilinguals, and English-dominantre (LEAP-Q). Standard deviations are presented in parentheses.


to a Korean person and English when speaking to anAmerican person. Though some reduction effects are sensi-tive to the presence of addressees (e.g. Arnold, Kahn, &Pancani, 2012; Rosa, Finch, Bergeson, & Arnold, 2015), rep-etition reduction has also been demonstrated in theabsence of an addressee (Lam & Watson, 2010, 2014).Furthermore, repetition reduction occurs even when theaddressee has changed (Bard & Aylett, 2004; Galati &Brennan, 2010; Meagher & Fowler, 2014), suggesting thatat least some aspect of repetition reduction is speakerdriven.

Displays were presented using MATLAB with thePsychophysics toolbox version 3 (Brainard, 1997; Kleiner,

Example 1.

Non-repeated, Stay, English: The book is shrinking . . . the apple is blinking.

Repeated, Stay, English: The apple is shrinking . . . the apple is blinking.

Non-repeated, Code-switch, English: 책이 줄어든다 . . . the apple is blinking.

(book shrinking)

Repeated, Code-switch, English: 사과가 줄어든다 . . . the apple is blinking.

(apple shrinking)

Brainard, & Pelli, 2007) and the CogToolbox for MATLAB(Fraundorf et al., 2014). Participants’ responses wererecorded at a frequency of 44,100 Hz using a SennheiserPC 360 microphone headset. The headset was used to con-trol the distance between speakers’ mouths and the micro-phone across trials, as distance can affect the recordedintensity of the sound wave.

Procedure

When participants first arrived, they completed an elec-tronic version of the LEAP-Q. Following the LEAP-Q, partic-ipants began the event description task.

At the beginning of the event description task, partici-pants were first shown a short video demonstrating the task.After watching the video, participants were visually pre-sented with the task instructions on the computer screen,first in English, and then in Korean. After reading the instruc-tions, participants completed practice trials from four differ-ent conditions of the experiment. Upon finishing the practice,participants completed 48 critical trials (without any fillertrials). The number of critical trials and the absence of fillertrials were based on previous work in the repetition reduc-tion literature (e.g., Lam & Watson, 2014).

At the start of a trial, two objects appeared on the com-puter screen followed by a picture of either a famousAmerican actress (Jennifer Aniston) or a famous Koreanactress (Son Yejin) to establish language mode (seeFig. 2). After three seconds, the picture of the actress disap-peared, and one of the objects began shrinking.Participants described this shrinking event2 in either

2 Participants were not explicitly told how to describe each event, thougha possible sentence structure was provided in the example video.

English or Korean, depending on which actress theysaw, and pressed a key to continue. Then, another pic-ture of either Jennifer Aniston or Son Yejin appeared,and one of the objects began blinking (the same twoobjects from before remained on the screen betweenevents to ensure consistency of referents). Participantsdescribed the blinking event and then pressed a key toend the trial. Sample sentences for each English-response condition are presented in Example 1. In eachsentence, the critical word is the noun of the secondutterance (underlined and bolded). Fig. 2 presents adepiction of the sequence of events in the RepeatedCodeswitch English condition.

Data analysis

Across all participants, 157 trials (6.8%) were excludedfrom analysis either because participants failed to namethe target object (4.9%) or because participants used thewrong language when describing the target object (1.9%).Because the events involved only shrinking and blinking,and participants were provided with an example video,there was very little variation in the carrier sentencesbetween trials for any particular participant. Across partici-pants, there was some variation in the carrier sentencesused in Korean, as some participants used the dictionary‘‘verb +다” form,whereas other participants used verb end-ings from the more formal ‘‘verb +ㅂ니다” form. In English,all participants used the verb ‘‘shrinking” as the first verb,but some participants used ‘‘flashing” as the second verbinstead of ‘‘blinking.”

Participants’ utterances in the event description taskwere hand-transcribed and annotated for prominenceanalysis in Praat, a speech analysis platform (Boersma &Weenink, 2007). Two fluent Korean–English bilingualswith early exposure to both languages completed the tran-scriptions. The transcribers marked the onset and offset ofthe target word as well as the onset and offset of the utter-ance. For Korean items, transcriptions for the target wordincluded the subject markers (가/이) which are attachedto the ends of the nouns in written Korean. Target wordonset and offset were identified by a combination of audi-tory perception and visual inspection of the spectrogram.Transcription duties were separated such that each persontranscribed utterances for different sets of participantswith an emphasis on within-transcriber consistency.Transcribers were given the following instructions formarking the onset and offset of target words.

3 For all reported analyses with unbalanced bilinguals, the patterns forthe main effect of repetition and the two-way repetition by code-switchinginteraction were the same as in models that only considered repetition,code-switching, and the repetition by code-switching interaction.

Fig. 2. Depiction of the sequence of events in the event description task from the Repeated Code-switch English condition. Speakers describe the eventsdepicted in Panel 3 (the shrinking event) and Panel 5 (the blinking event).


� For word boundaries that occur at a transition betweena vowel and consonant, set the boundary at the end ofthe pitch period coinciding with a change in amplitudevisible in the spectrogram.

� For word boundaries that occur at a vowel–vowel tran-sition, mark the boundary at the midpoint of the voweltransition in the spectrogram and verify perceptually bylistening to the sound file before and after the boundarymarker.

� When the target word is utterance initial, mark the tar-get word onset at the beginning of the visible change inintensity in the spectrogram from silence to the onset ofthe word and verify through auditory perception.

For Korean target words, the onset was alwaysutterance-initial while the offset always coincided with avowel–consonant transition. For English target words, theonset coincided with a vowel–consonant transition whenthe participant produced a determiner and the target wordbegan with a consonant onset (�87.5% of English targets).For English target word offsets, the offset occurred at avowel–consonant boundary for all target words that endedin a coda consonant (�75% of English targets).

As is typical in prominence analysis, we present the promi-nence data both in terms of duration (e.g., Fowler, 1988;Fowler & Housum, 1987; Galati & Brennan, 2010; Kahn &Arnold, 2012) and intensity (e.g., Kochanski et al., 2005; Lam& Watson, 2010, 2014; Watson et al., 2008) because thesemetrics may be sensitive to different aspects of reduction.Lam (2012, see also Lam & Watson, 2014) argues that inten-sity may be more sensitive to differences at the concept levelof production while duration may be more sensitive to differ-ences at the word and phoneme levels. This would lead to theprediction that intensity is more likely to show a main effectof repetition while duration is more likely to show an interac-tion between repetition and code-switching.

Analyses on duration were conducted on both the rawduration (ms) for the target word (i.e., the noun used inthe second utterance) and the percentage of the utterancelength accounted for by the target word (i.e., noun dura-tion/total utterance duration), which we will refer to as‘‘utterance proportion.” This second measure of durationprovides an estimate of how prominent a target word iswithin an utterance and accounts for inter-speaker vari-ability in speech duration. If repetition leads to reductionof an entire phrase, utterance proportions may not capturethis change, but if repetition affects the target word itself,utterance proportion may be a more sensitive measure oftarget word reduction. Though prominence differencestypically manifest on the stressed syllable of a word inEnglish, in Korean, prominence is a property of the accen-tual phrase (e.g., Jun, 1993, 1998), which may include onlya single word or multiple words (including a full sentence)

depending upon speech rate and word length. When theaccentual phrase includes the entire sentence, utteranceproportions may be less sensitive to this change.

As with duration, analyses on intensity examined bothraw and relative measures. Analyses on intensity were con-ducted on the average intensity of the target word measuredin decibels SPL (sound pressure level) and the relative inten-sity of the target wordmeasured as the percentage differencein intensity from the target word compared to the intensityof the rest of the utterance (i.e., [target intensity–utteranceintensity]/utterance intensity). As with relative duration, rel-ative intensity provides a measure of how prominent theword is within the utterance.

Initial analyses were conducted on all participants withrepetition, code-switching, and the interaction between rep-etition and code-switching as fixed effects of interest. Wealso included response-language (Korean vs. English) andlanguage-balance (i.e., balanced vs. unbalanced) as controlvariables, but do not report their effects as the initial analysesare focused on whether repetition reduction can occur forconcept-level repetition in the absence of word-level repeti-tion. Contrasts for all fixed effects were sum coded.

Following the analyses with all participants, we con-ducted follow-up analyses examining the effect of lan-guage balance (e.g., balanced bilinguals vs. unbalancedbilinguals) on repetition reduction. For follow-up analyseswith balanced bilinguals, we considered fixed effects ofrepetition and code-switching, as well as their interactionswith response language modeled as a control variable. Forunbalanced bilinguals, we considered fixed effects of repe-tition, code-switching, response-dominance (i.e., dominantvs. non-dominant language), and participant type (i.e.,Korean-dominant bilingual vs. English-dominant bilingual)as well as their interactions. Unlike in the analyses withbalanced bilinguals, for unbalanced bilinguals, response-dominance and participant type were modeled as fixedeffects because unbalanced bilinguals would be expectedto show different patterns of prominence across theirtwo languages.3 Note that for unbalanced bilinguals wehave recoded the ‘‘response-language” factor as ‘‘response-dominance.” This is because the dominance of the responselanguage is more important theoretically for our researchquestions than the identity of the language itself. As in themain analyses, contrasts for all fixed effects were sum coded.

Analyses were conducted using multilevel modeling withrandom slopes and intercepts for subjects and items (Barr,Levy, Scheepers, & Tily, 2013). As with analysis of variance


(ANOVA), multilevel models are able to account for variancefrom subjects and items; however this method has the addi-tional benefit of being able to account for multiple randomfactors simultaneously (Baayen, Davidson, & Bates, 2008).For all analyses, we used the maximally converging randomeffects structure for subjects and items. If the model did notconverge, we removed random slope terms from the modelstarting with the highest order term that accounted for theleast variance in the model. Tables 2–4 present the randomterms that were included or excluded in each of the maxi-mally converging models for analyses with all participants(Table 2), analyses with balanced bilinguals (Table 3), andanalyses with unbalanced bilinguals (Table 4).

The p-value estimates were obtained using normalapproximation by assuming that the t distributionapproached a z distribution given our number of observa-tions. As this study is focused on the effects of repetitionwithin and across languages, we report the main effect ofrepetition and the interaction between repetition andcode-switching for all analyses. We also report all signifi-cant interactions that include the factor of repetition. Forsignificant interactions (p < .05) involving repetition,follow-up analyses were conducted to examine the pat-terns of the interactions by separating analyses for eachlevel of a variable, while retaining all other fixed effectsthat do not include the separated variable. A completeset of results tables is included in Appendix B.

Results

Concept repetition in the presence and absence of wordrepetition

DurationThe means for raw duration across all participants are pre-

sented in Fig. 3A. For raw duration, there was a significanttwo-way interaction between repetition and code-switching(b =�54.63, SE = 10.13, t =�5.39, p < .001). Follow-up analy-ses examining the effect of repetition across stay and code-switch trials demonstrated that repeated targets were pro-duced with shorter durations than non-repeated targets inthe stay condition (b = 32.50, SE = 5.08, t = 6.39, p < .001) butnot in the code-switch condition (b =�11.25, SE = 7.49,t =�1.50, p = .133). There was also a main effect of repetition:repeated targets were produced with shorter durations thannon-repeated targets (b = 9.46, SE = 4.42, t = 2.13, p < .033).

The means for utterance proportion across all partici-pants are presented in Fig. 4A. The patterns for utteranceproportion were similar to raw durations. There was a sig-nificant two-way interaction between repetition and code-switching (b = �1.70, SE = 0.40, t = �4.26, p < .001). Whenstay and code-switch trials were analyzed separately,repeated targets were produced with smaller utteranceproportions than non-repeated targets in the stay condi-tion (b = 1.75, SE = 0.27, t = 6.50, p < .001), but not in thecode-switch condition (b = 0.07, SE = 0.32, t = 0.23,p = .818). In addition, there was a significant main effectof repetition where repeated targets were produced withsmaller utterance proportions than non-repeated targets(b = 0.96, SE = 0.22, t = 4.32, p < .001).

IntensityFig. 5A presents the means for raw intensity across all

participants. For raw intensity, there was a significantmain effect of repetition in which repeated targets wereproduced with lower intensity that non-repeated targets(b = 0.24, SE = 0.11, t = 2.14, p = .032). The interactionbetween repetition and code-switching was not significant(b = �0.30, SE = 0.26, t = �1.15, p = .250).

The means for relative intensity across all participantsare presented in Fig. 6A. For relative intensity, there wasa significant two-way interaction between repetition andcode-switching (b = �1.09, SE = 0.52, t = �2.10, p = .036).Follow-up analyses separating stay and code-switch trialsdemonstrated that repeated targets were produced withlower relative intensity than non-repeated targets on staytrials (b = 1.02, SE = 0.41, t = 2.50, p = .013) but not oncode-switch trials (b = �0.12, SE = 0.32, t = �0.37,p = .707). In addition, there was a significant main effectof repetition where repeated targets were produced withlower relative intensity than non-repeated targets(b = 0.75, SE = 0.27, t = 2.82, p = .005).

Repetition reduction and language balance

When language-balance was added to the models thatexamined the locus of repetition reduction across all par-ticipants, the results indicated that language-balancemediated repetition reduction (see Appendix B).Specifically, there were significant two-way interactionsbetween repetition and language-balance for raw duration(b = 17.61, SE = 6.85, t = 2.57, p = .010) and utterance pro-portion (b = 1.00, SE = 0.38, t = 2.62, p = .008), and therewas a significant three-way interaction between repeti-tion, code-switching, and language-balance for relativeintensity (b = 2.18, SE = 1.00, t = 2.17, p = .030). In the nextsection we present follow-up analyses separating balancedand unbalanced bilinguals to examine the effects oflanguage-balance on repetition reduction.

Balanced bilinguals

DurationThe means for raw duration in balanced bilinguals are

presented in Fig. 3B. For raw duration, there was a signifi-cant main effect of repetition where repeated targets wereproduced with shorter duration than non-repeated targets(b = 29.10, SE = 5.82, t = 5.00, p < .001). The interaction ofrepetition by code-switching was not significant(b = �22.80, SE = 13.70, t = �1.66, p = .096).

The means for utterance proportion are presented inFig. 4B. The pattern for utterance proportion was similarto the pattern for raw duration. For utterance proportion,there was a main effect of repetition where repeated tar-gets were produced with smaller utterance proportionsthan non-repeated targets (b = 1.53, SE = 0.29, t = 5.29,p < .001). The interaction between repetition and code-switching was not significant (b = �0.64, SE = 0.59,t = �1.08, p = .281).

Table 2Maximally converging random effects structure for models with all participants.

Subject random effects Item random effects

Rawduration

Utteranceproportion

Rawintensity

Relativeintensity

Rawduration

Utteranceproportion

Rawintensity

Relativeintensity

(Intercept) � � � � � � � �Repetition � � � � � � � �Codeswitch � � � � � � � �Language � � � �Balance � � � �Repetition:Codeswitch � � � � � � � �Repetition:Language – � – –Codeswitch:Language � � � �Repetition:Balance � � – �Codeswitch:Balance – – – –Repetition:Codeswitch:

Language– – – –

Repetition:Codeswitch:Balance

– – – –

Note: Asterisks indicate the slope and intercept terms that were included in the maximally converging model for each metric. Dashes indicate slope termsthat were excluded from the maximally converging model for each metric. Empty boxes represent slope terms that are not theoretically possible.

Table 3Maximally converging random effects structures for models with balanced bilinguals.


Rawduration

Utteranceproportion

Rawintensity

Relativeintensity

Rawduration

Utteranceproportion

Rawintensity

Relativeintensity

(Intercept) � � � � � � � �Repetition � � � � � � � �Codeswitch � � � � � � � �Language � � � �Repetition:Codeswitch � – � � � – � �Repetition:Language � – � �Codeswitch:Language � – � �Repetition:Codeswitch:

Language– – � �


Table 4Maximally converging random effects structures for models with unbalanced bilinguals.


Rawduration

Utteranceproportion

Rawintensity

Relativeintensity

Rawduration

Utteranceproportion

Rawintensity

Relativeintensity

(Intercept) � � � � � � � �Repetition � � � � � � � –Codeswitch � � � � � � � �Language � � � �PartType � � � �Repetition:Codeswitch � – – – � – – –Repetition:DomLang � – – –Codeswitch:DomLang � – – –Repetition:PartType – – – –Codeswitch:PartType – – – –Repetition:Codeswitch:

DomLang– – – –

Repetition:Codeswitch:PartType

– – – –




IntensityThe means for raw intensity in balanced bilinguals are

presented in Fig. 5B. Unlike with raw duration and utter-ance proportion, raw intensity showed no clear patterns.The interaction between repetition and code-switchingwas not significant (b = 0.12, SE = 0.44, t = 0.29, p = .772)and the main effect of repetition was also not significant(b = 0.28, SE = 0.27, t = 1.04, p = .299).

Fig. 3. Average raw durations for words produced by all participantscombined (Panel A), balanced bilinguals (Panel B), and unbalancedbilinguals (Panel C). The left side of each chart depicts the main effectof repetition collapsed across Stay and Code-switch trials. Asterisksindicate significant effects (p < .05).

The means for relative intensity in balanced bilingualsare presented in Fig. 6B. As with raw intensity, the analysesfor relative intensity yielded no significant effects or repe-tition (b = 0.37, SE = 0.61, t = 0.60, p = .547) nor a significantinteraction between repetition and code-switching(b = 0.45, SE = 0.91, t = 0.49, p = .622).

Fig. 4. Average utterance proportions for words produced by all partic-ipants combined (Panel A), balanced bilinguals (Panel B), and unbalancedbilinguals (Panel C). The left side of each chart depicts the main effect ofrepetition collapsed across Stay and Code-switch trials. Asterisks indicatesignificant effects (p < .05).

Fig. 5. Average raw intensities for words produced by all participantscombined (Panel A), balanced bilinguals (Panel B), and unbalancedbilinguals (Panel C). The left side of each chart depicts the main effectof repetition collapsed across Stay and Code-switch trials. Asterisksindicate significant effects (p < .05).

Fig. 6. Average relative intensities for words produced by all participantscombined (Panel A), balanced bilinguals (Panel B), and unbalancedbilinguals (Panel C). The left side of each chart depicts the main effectof repetition collapsed across Stay and Code-switch trials. Asterisksindicate significant effects (p < .05).


Unbalanced bilinguals

DurationMeans for raw duration in unbalanced bilinguals are

presented in Fig. 3C. Unlike in balanced bilinguals, repeti-tion interacted with code-switching in unbalanced bilin-guals. For raw duration there was a significant three-way

interaction between repetition, code-switching, andresponse-dominance (b = �45.42, SE = 15.47, t = �2.94,p = .003). When responding in the non-dominant language,there was a significant repetition by code-switching inter-action (b = �49.90, SE = 13.23, t = �3.77, p < .001). Thoughrepeated targets were produced with shorter durations


than non-repeated targets in the stay condition (b = 29.51,SE = 10.16, t = 2.90, p = .004), repeated targets were pro-duced with longer durations than non-repeated targets inthe switch condition (b = �42.03, SE = 14.60, t = �2.88,p = .004). When responding in the dominant language,the interaction between repetition and code-switchingwas not significant (b = �21.26, SE = 12.63, t = �1.68,p = .092).

There was also a significant two-way interactionbetween repetition and code-switching (b = �62.75,SE = 14.95, t = �4.20, p < .001). Repeated targets were pro-duced with shorter duration than non-repeated targets inthe stay condition (b = 30.52, SE = 7.29, t = 4.18, p < .001),but were produced with longer durations than non-repeated targets in the code-switch condition(b = �22.60, SE = 9.10, t = �2.49, p = .013). The main effectof repetition was not significant (b = 2.68, SE = 4.77,t = 0.56, p = .574).

For utterance proportion, there was a significant three-way interaction (b = �2.19, SE = 0.85, t = �2.57, p = .010)between repetition, code-switching, and response-dominance (see Fig. 4C). When responding in their non-dominant language, there was a significant interactionbetween repetition and code-switching (b = �2.86,SE = 0.70, t = �4.10, p < .001), in which repeated targetswere produced with smaller utterance proportions thannon-repeated targets in the stay condition (b = 1.81,SE = 0.51, t = 3.52, p < .001), but with larger utterance pro-portions than non-repeated targets in the switch condition(b = �1.53, SE = 0.57, t = �2.69, p = .007). When unbalancedbilinguals responded in their dominant language, the inter-action between repetition and code-switching was not sig-nificant (b = �0.99, SE = 0.57, t = �1.75, p = .081).

There was also a significant three-way interactionbetween repetition, response-dominance, and participanttype (b = �1.77, SE = 0.86, t = �2.06, p = .039). Whenresponding in the dominant language, there was a signifi-cant interaction between repetition and participant type(b = 1.41, SE = 0.68, t = 2.09, p = .037). English-dominantbilinguals demonstrated amain effect of repetition inwhichrepeated targetswere producedwith smaller utterance pro-portion compared to non-repeated targets (b = 1.64,

Fig. 7. Average utterance proportions in the dominant language for targets produKorean-dominant bilinguals (Panel B). The left side of each chart depicts the mainindicate significant effects (p < .05).

SE = 0.48 t = 3.45, p < .001), while Korean-dominant bilin-guals did not show a main effect (b = 0.23, SE = 0.48,t = 0.48, p = .635) of repetition (see Fig. 7A and B). Whenresponding in the non-dominant language, the interactionbetween repetition and participant type was not significant(b = �0.75, SE = 0.71, t = �1.06, p = .291).

Utterance proportions in unbalanced bilinguals alsoshowed a significant two-way interaction between repeti-tion and code-switching (b = �1.77, SE = 0.43, t = �4.13,p < .001). Repeated targets were produced with smallerutterance proportions than non-repeated targets in thestay condition (b = 1.58, SE = 0.35, t = 4.47, p < .001), butnot in the code-switch condition (b = �0.55, SE = 0.37,t = �1.49, p = .135). Finally, there was a significant maineffect of repetition on utterance proportion (b = 0.55,SE = 0.23, t = 2.43, p = .015) in which repeated targets wereproduced with smaller utterance proportions than non-repeated targets. Given the interaction between repetitionand participant type, the main effect of repetition waslikely driven by English-dominant bilinguals.

IntensityThere was a significant interaction between repetition

and code-switching (b = �0.58, SE = 0.25, t = �2.30,p = .021) in unbalanced bilinguals (see Fig. 5C). Repeatedtargets were produced with lower raw intensity thannon-repeated targets in the stay condition (b = 0.47,SE = 0.17, t = 2.73, p = .006) but not in the code-switch con-dition (b = �0.11, SE = 0.20, t = �0.57, p = .569). The maineffect of repetition was not significant (b = 0.21, SE = 0.13,t = 1.66, p = .097).

The means for relative intensity in unbalanced bilin-guals are presented in Fig. 6C. For relative intensity, therewas a significant interaction between repetition andcode-switch (b = �1.82, SE = 0.51, t = �3.58, p < .001) inwhich repeated targets were produced with lower relativeintensity than non-repeated targets in the stay condition(b = 1.38, SE = 0.46, t = 2.97, p = .003), but not in the code-switch condition (b = �0.59, SE = 0.43, t = �1.37, p = .171).The main effect of repetition was not significant for relativeintensity (b = 0.42, SE = 0.30, t = 1.38, p = .167).

ced by English-dominant bilinguals (Panel A) and for targets produced byeffect of repetition collapsed across Stay and Code-switch trials. Asterisks


Discussion

When bilingual Korean–English speakers performed atask in which they repeatedly mentioned targets withand without a language switch, repeated targets werereduced when the language did not change, but were notreduced when the language changed. This pattern wasseen for raw duration, utterance proportion, and relativeintensity. For raw intensity, repetition led to reductionwhether or not there was a language switch, signalingpotential differences in how duration and intensity conveyprominence in an utterance (see Lam & Watson, 2010;Watson et al., 2008).

We also saw different patterns of repetition reductionfor balanced and unbalanced bilinguals. For raw durationand utterance proportion, balanced bilinguals demon-strated repetition reduction whether or not the word itselfwas repeated, suggesting that concept-level repetition canlead to prominence reduction even in the absence of word-level repetition. In contrast, unbalanced bilinguals did notshow effects of concept-level repetition.

A significant three-way interaction between repetition,code-switching and language-balance for relative intensitysuggests that balanced and unbalanced bilinguals alsorespond differently to word-level repetition. While unbal-anced bilinguals demonstrated repetition reduction forword-level repetition, balancedbilinguals did not show rep-etition reduction for relative intensity. Conversely, the lackof a three-way interaction for raw duration or utteranceproportion suggests that for duration, balanced bilingualsand unbalanced bilinguals may exhibit similar patterns ofrepetition reduction for word-level repetition. However,follow-up analyses indicated that word-level repetitiononly affects repetition reduction in unbalanced bilinguals.

Though unbalanced bilinguals demonstratedword-levelrepetition reduction in general, there were different pat-terns across the dominant vs. non-dominant language forraw duration and utterance proportion.While repeated tar-gets produced in the non-dominant language were reducedwhen the languagewas also repeated, theywere lengthenedwhen therewas a language switch; this pattern did not holdin the non-dominant language. For raw and relative inten-sity, unbalanced bilinguals reduced repeated targets in thesame language, but not when the language changed.Unlike duration, intensity was unaffected by responselanguage dominance. Moreover, Korean-dominant bilin-guals and English-dominant bilinguals displayed slightlydifferent patterns of repetition reduction for utterance pro-portion. When production was in their dominant language,English-dominant bilinguals exhibited repetition reductionfor concept repetition, but Korean-dominant bilinguals didnot. When responding in the non-dominant language,neither English-dominant nor Korean-dominant bilingualsshowed any effects of concept repetition.

Locus of repetition reduction

The results of our study suggest that not all reductioncan be attributed to concept level repetition. Because con-cepts are shared across languages (e.g., Costa, 2005; Kroll &

Stewart, 1994), if repetition reduction were situated onlyat the concept level, then repetition reduction should bepresent whether or not the word itself was repeated.While the main effect of repetition for intensity was con-sistent with this prediction, the observed interactionbetween repetition and code-switching in the other threemetrics suggests that repetition reduction must be at leastpartly influenced by word-level repetition. In addition,follow-up comparisons between balanced and unbalancedbilinguals indicated that the locus of repetition reduction isaffected by the connection strength between words andconcepts.

Although not all repetition reduction can be attributedto concept-level repetition, the results for balanced bilin-guals suggest that concepts do play a role in repetitionreduction. Even when the language, and thus the word,had changed, balanced bilinguals reduced duration forrepeated targets. This suggests that repetition reductioncan exist without word repetition. Given this pattern, it ispossible that repetition reduction manifests through therepeated access of a particular concept. In this sense, prioractivation caused the concept to be more accessible, lead-ing to faster subsequent retrieval of the concept.

Nevertheless, the patterns of reduction in unbalancedbilinguals are most consistent with word-level repetitionreduction. In support of word-mediated reduction, unbal-anced bilinguals demonstrated repetition reduction forrepeated targets in the stay trials (concept and word rep-etition), but not in the code-switch trials (concept repeti-tion only). Furthermore, for measures of duration, whenunbalanced bilinguals switched from the dominant lan-guage to the less dominant language in the concept rep-etition only condition, rather than being reduced inprominence, the repeated targets were instead producedwith greater prominence (longer raw durations and largerutterance proportions) than non-repeated targets. Thispattern may be indicative of a greater cost of switchingfrom a word in the dominant language to a word in thenon-dominant language (e.g., Kroll et al., 2008) whenrepeatedly mentioning the same target. These results sug-gest that repetition reduction is also influenced by theaccessibility of the words used to describe a particularreferent.

When data from balanced and unbalanced bilinguals areconsidered together, the pattern of results suggests thatrepetition reduction may be influenced by both conceptand word repetition. Though repetition reduction mayoccur in the absence of word repetition in balanced bilin-guals, unbalanced bilinguals demonstrated repetitionreduction only in the presence of both concept and wordrepetition. One explanation for the different patterns in bal-anced vs. unbalanced bilinguals may be that repetitionreduction is mediated by the link between the conceptand the word (i.e., the step between determining the con-cept and retrieving the appropriate word). A number of the-ories suggest that concepts may activate multiple words sothat words that are highly related to target words alsoreceive some activation (Costa et al., 2000; Dell, 1986;Dell & O’Seaghdha, 1991; Levelt et al., 1999; Peterson &Savoy, 1998). In the case of translation equivalents, concep-tual information is nearly identical. While this could lead to


translation equivalents having similar levels of activation astarget words for balanced bilinguals, concepts may activatewords in each language to different degrees in unbalancedbilinguals because unbalanced bilinguals have strongerlinks between concepts and words in their dominant lan-guage than in their less dominant language (Kroll &Stewart, 1994). When a concept activates multiple words(as in balanced bilinguals), there may be repetition reduc-tion without explicit word repetition because the non-selected words still receive some activation. If the conceptpreferentially activates a single word (as in unbalancedbilinguals), there may be no effect of concept repetition inthe absence of word repetition because the non-selectedwords did not receive any activation. As a result, balancedbilinguals reduce repeated targets both with and withoutword repetitionwhereas unbalanced bilinguals only reducerepeated targets when there was word repetition.

Another possible explanation for our results is thatrepetition reduction exists for repeated concepts, but thiseffect was obscured because unbalanced bilinguals havedifficulty inhibiting the previously-used word due to therecency and frequency of use biases (e.g., Brennan &Clark, 1996; Garrod & Anderson, 1987; Garrod &Doherty, 1994). Balanced bilinguals would not have thesefrequency biases because the words are balanced acrossthe two languages (Costa et al., 2000; Gollan & Ferreira,2009). The previously-used word may be a salient com-petitor leading to slower retrieval of the target word,which may offset any effect of repeated concepts. Thisexplanation is consistent with the fact that when unbal-anced bilinguals switched from the dominant languageto the non-dominant language, repeated targets wereactually produced with greater prominence rather thanreduced prominence. Because unbalanced bilingualsrequire more effort to inhibit words in their dominantlanguage than in their non-dominant language (Krollet al., 2008), if production of the first utterance was intheir dominant language, the target word from the firstutterance may be more accessible than the word in thenon-dominant language. According to accessibility theo-ries of reduction (e.g., Bell et al., 2009; Kahn & Arnold,2012; Lam & Watson, 2010), differences in the accessibil-ity of words should lead to differences in prominencesuch that accessible words are produced with less promi-nence whereas less accessible words are produced withgreater prominence.

Difficulty in suppressing a previous word may alsoexplain the differences between Korean-dominant bilin-guals and English-dominant bilinguals with respect toutterance proportion. Whereas both groups reducedrepeated targets in the stay condition, in the code-switchcondition, English-dominant bilinguals produced repeatedtargets with smaller utterance proportions if they wereresponding in their dominant language (i.e., English), whileKorean-dominant bilinguals did not show a repetition byresponse-language interaction. Given that the study wasconducted in an English-speaking country and context,English words may be more accessible than Korean words.When considered together with their already existing pref-erence to use English, the English-dominant bilinguals

should have less difficulty suppressing the Korean wordfor the English translation, which could allow for repetitionreduction in the absence of word repetition.

Our results for balanced bilinguals differed fromMonsell et al. (1992), for which word repetition led toreduction, but conceptual repetition did not. There are afew possible explanations for the discrepancy betweentheir results and ours. First, while our balanced bilingualsshowed an effect of concept repetition, our unbalancedbilinguals showed an effect that was similar to Monsellet al. (1992). For unbalanced bilinguals, repetition reduc-tion typically occurred only when the language was thesame across productions (i.e., word repetition). Monsellet al. (1992) presented no data on language balanceamong their bilinguals. It is possible that if they had sep-arated their analysis by language balance, they may haveshown conceptual repetition effects for balanced bilin-guals. Second, the time scales of the two studies were dif-ferent. In Monsell et al. (1992), concept repetitionoccurred over a large time scale and in a very differenttype of task across the two productions. In our study,the repeated target was mentioned in the immediatelypreceding sentence. This shorter time difference is bettersuited to a study of reference production because previ-ous referring expressions are still available. Finally, unlikein Monsell et al. (1992), repeated targets in our manipu-lation were presented in the same visual context bothwithin and across languages. Although speakers usedtwo different words in the repeated-target code-switchcondition, both words referred to the same object, andthe object was visible on the screen between the twoevents. In Monsell et al. (1992), repetition primes camefrom sentence completions or definitions in a separatetask. The change of both referent and context may haveled different concepts to be activated across the two tasksbecause the meaning of a word changes depending uponits context (Olson, 1970).

Conclusions

In conclusion, our experiment suggests that repetitionreduction is linked to repetition at both the concept leveland the word level. Across a mixed group of bilingualspeakers, participants reduced repeated targets onlywhen the word itself was repeated, suggesting that repe-tition of concepts is not sufficient for repetition reduction.However, a more nuanced conclusion is reached whenseparating analyses based upon language balance. Whileunbalanced bilinguals showed only word-mediated repe-tition reduction, balanced bilinguals showed reductionwhether or not the word itself was repeated. This sug-gests that concept repetition can affect repetition reduc-tion in the absence of word repetition when stronglinks exist between concepts and words in both lan-guages (as is the case for balanced bilinguals). We suggestthat the connections between concepts and words withina speaker’s language system determine whether or notrepetition reduction can occur in the absence of wordrepetition.

Appendix A

Individual participant language profiles

Participant Age of acquisitionKorean

Age of acquisitionEnglish

Koreanproficiency

Englishproficiency

Languagebalance

CP1 0 8 10 8.5 KoreanCP2 0 1 10 8 KoreanCP3 1 1 10 6 KoreanCP4 0 16 10 7.5 KoreanCP5 1 12 10 10 BalancedCP6 1 1 9 7.5 KoreanCP7 0 10 10 8 KoreanCP8 1 10 10 9 BalancedCP9 1 2 9 10 BalancedCP10 0 4 10 8 KoreanCP11 0 10 10 10 BalancedCP12 0 3 8.5 10 EnglishCP13 2 8 9.5 10 BalancedCP14 2 4 7 8.5 EnglishCP15 0 10 10 8.5 KoreanCP16 0 4 9.5 5 KoreanCP17 0 7 10 7.5 KoreanCP18 0 0 8 8 BalancedCP19 0 8 8.5 10 EnglishCP20 0 8 10 8.5 KoreanCP21 1 7 9 8 BalancedCP22 0 5 10 9.5 BalancedCP23 0 7 9 9.5 BalancedCP24 0 7 10 8 KoreanCP25 0 10 9 8 BalancedCP26 0 4 10 10 BalancedCP27 0 3 9.5 8 KoreanCP28 2 1 9 10 BalancedCP29 0 6 7.5 10 EnglishCP30 1 1 10 10 BalancedCP31 0 10 9.5 8 KoreanCP32 0 2 10 10 BalancedCP33 0 3 9 8.5 BalancedCP34 0 6 8.5 9.5 EnglishCP35 0 5 8.5 10 EnglishCP36 1 3 9 10 BalancedCP37 0 0 8 10 EnglishCP39 3 8 7.5 10 EnglishCP41 2 6 6.5 10 EnglishCP38 0 12 10 8.5 KoreanCP40 0 3 8 10 EnglishCP42 2 4 8.5 10 EnglishCP43 1 4 10 7 KoreanCP44 1 4 8 10 EnglishCP45 1 3 7.5 9 EnglishCP46 0 10 8.5 10 EnglishCP47 0 4 7 10 EnglishCP48 0 9 8.5 10 English


Appendix B: Analysis results tables

In this appendix, we present the full results tables for all fixed effects in each of our analyses on all participants, balancedbilinguals, and unbalanced bilinguals.

1. Results tables for analyses on all participants (models excluding language-balance).2. Results tables for analyses on all participants (models including language-balance).3. Results tables for analyses on balanced bilinguals.4. Results tables for analyses on unbalanced bilinguals.

Results tables for analyses on all participants (models excluding language-balance)

1a. Raw duration (All Participants, model excluding language-balance).

Estimate Std. error t-value p

(Intercept) 533.15 17.42 30.61 <1e�03Repetition 9.46 4.42 2.13 .033Codeswitch 13.84 3.93 3.52 <1e�03Repetition:Codeswitch �54.63 10.13 �5.39 <1e�03

Note: For the factor ‘‘Repetition,” the repeated level is coded negative while the non-repeated level is coded positive. For the factor ‘‘Codeswitch,” the staylevel is coded negative while the code-switch level is coded positive.

1b. Utterance proportion (all participants, model excluding language-balance).


(Intercept) 44.67 0.87 51.26 <1e�03Repetition 0.96 0.22 4.32 <1e�03Codeswitch 0.90 0.18 4.90 <1e�03Repetition:Codeswitch �1.70 0.40 �4.26 <1e�03


1c. Raw intensity (All Participants, model excluding language-balance).


(Intercept) 62.27 0.67 92.37 <1e�03Repetition 0.24 0.11 2.14 .032Codeswitch 0.50 0.12 4.15 <1e�03Repetition:Codeswitch �0.30 0.26 �1.15 .250


1d. Relative intensity (All Participants, model excluding language-balance).


(Intercept) 5.80 0.45 13.00 <1e�03Repetition 0.75 0.27 2.82 .005Codeswitch 1.20 0.23 5.14 <1e�03Repetition:Codeswitch �1.09 0.52 �2.10 .036



Results tables for analyses on all participants (models including language-balance)

2a. Raw duration (all participants, model including language-balance).


(Intercept) 530.94 17.68 30.02 <1e�03Repetition 8.77 4.37 2.00 .045Codeswitch 13.55 3.93 3.45 <1e�03Balance �11.54 15.21 �0.76 .448Repetition:Codeswitch �52.90 9.55 �5.54 <1e�03Repetition:Balance 17.61 6.85 2.57 .010Codeswitch:Balance 14.56 7.05 2.06 .039Repetition:Codeswitch:Balance 18.41 13.68 1.35 .178

Note: For the factor ‘‘Repetition,” the repeated level is coded negative while the non-repeated level is coded positive. For the factor ‘‘Codeswitch,” the staylevel is coded negative while the code-switch level is coded positive. For the factor ‘‘Balance,” the unbalanced bilingual level is coded negative while thebalanced bilingual level is coded positive.

2b. Utterance proportion (All Participants, model including language-balance).


(Intercept) 44.45 0.88 50.50 <1e�03Repetition 0.95 0.21 4.40 <1e�03Codeswitch 0.84 0.17 4.88 <1e�03Balance �0.11 0.95 �0.12 .906Repetition:Codeswitch �1.56 0.37 �4.17 <1e�03Repetition:Balance 0.99 0.38 2.61 .009Codeswitch:Balance 0.63 0.37 1.70 .089Repetition:Codeswitch:Balance 0.95 0.74 1.29 .197


2c. Raw intensity (All Participants, model including language-balance).


(Intercept) 62.24 0.67 93.53 <1e�03Repetition 0.24 0.11 2.16 .031Codeswitch 0.50 0.12 4.14 <1e�03Balance 2.27 1.39 1.62 .103Repetition:Codeswitch �0.29 0.21 �1.38 .162Repetition:Balance �0.06 0.23 �0.24 .807Codeswitch:Balance �0.36 0.25 �1.44 .150Repetition:Codeswitch:Balance 0.74 0.45 1.63 .101


2d. Relative intensity (All Participants, model including language-balance).


(Intercept) 5.78 0.45 12.96 <1e�03Repetition 0.75 0.27 2.81 .005Codeswitch 1.19 0.23 5.11 <1e�03

(continued on next page)


Results tables for analyses on all participants (models including language-balance) (continued)


Balance �0.49 0.79 �0.63 .530Repetition:Codeswitch �1.15 0.52 �2.21 .027Repetition:Balance �0.05 0.60 �0.08 .933Codeswitch:Balance �0.13 0.53 �0.24 .808Repetition:Codeswitch:Balance 2.18 1.00 2.17 .030


Results tables for analyses on balanced bilinguals

3a. Raw duration (Balanced Bilinguals).


(Intercept) 491.28 18.86 26.05 <1e�03Repetition 29.10 5.82 5.00 <1e�03Codeswitch 17.69 7.03 2.52 .012Repetition:Codeswitch �22.80 13.70 �1.66 .096


3b. Utterance proportion (Balanced Bilinguals).


(Intercept) 43.51 1.08 40.45 <1e�03Repetition 1.53 0.29 5.28 <1e�03Codeswitch 0.80 0.33 2.44 .015Repetition:Codeswitch �0.64 0.59 �1.08 .281


3c. Raw intensity (Balanced Bilinguals).


(Intercept) 62.90 1.11 56.70 <1e�03Repetition 0.28 0.27 1.04 .299Codeswitch 0.21 0.21 0.99 .320Repetition:Codeswitch 0.13 0.44 0.29 .772


3d. Relative intensity (balanced bilinguals).


(Intercept) 4.44 0.66 6.70 <1e�03Repetition 0.37 0.61 0.60 .547Codeswitch 0.87 0.40 2.18 .029Repetition:Codeswitch 0.45 0.91 0.49 .622



Results tables for analyses on unbalanced bilinguals

4a. Raw duration (unbalanced bilinguals).


(Intercept) 531.76 19.46 27.32 <1e�03Repetition 2.68 4.77 0.56 .574Codeswitch 8.54 4.91 1.74 .082DomLang 24.11 11.38 2.12 .034PartType 10.83 20.80 0.52 .603Repetition:Codeswitch �62.75 14.95 �4.20 <1e�03Repetition:DomLang �10.23 9.55 �1.07 .284Codeswitch:DomLang 4.11 10.03 0.41 .682Repetition:PartType 7.49 8.24 0.91 .363Codeswitch:PartType 0.45 8.96 0.05 .960DomLang:PartType �3.66 69.47 �0.05 .958Repetition:Codeswitch:DomLang �45.42 15.47 �2.94 .003Repetition:Codeswitch:PartType �15.51 20.49 �0.76 .449Repetition:DomLang:PartType �7.70 21.38 �0.36 .719Codeswitch:DomLang:PartType 43.14 21.58 2.00 .046Repetition:Codeswitch:DomLang:PartType 1.41 53.42 0.03 .979

Note: For the factor ‘‘Repetition,” the repeated level is coded negative while the non-repeated level is coded positive. For the factor ‘‘Codeswitch,” the staylevel is coded negative while the code-switch level is coded positive. For the factor ‘‘DomLang”, the dominant language level is coded negative while thenon-dominant language level is coded positive. For the factor ‘‘PartType,” the Korean-dominant level is coded negative while the English-dominant level iscoded positive.

4b. Utterance proportion (unbalanced bilinguals).


(Intercept) 44.38 0.85 52.04 <1e�03Repetition 0.55 0.23 2.43 .015Codeswitch 0.81 0.22 3.68 .000DomLang 0.79 0.79 1.00 .318PartType �1.76 1.05 �1.67 .095Repetition:Codeswitch �1.77 0.43 �4.13 <1e�03Repetition:DomLang �0.52 0.42 �1.22 .224Codeswitch:DomLang 0.02 0.43 0.06 .955Repetition:PartType 0.27 0.45 0.61 .543Codeswitch:PartType 0.14 0.43 0.33 .743DomLang:PartType �20.16 3.10 �6.50 <1e�03Repetition:Codeswitch:DomLang �2.19 0.85 �2.57 .010Repetition:Codeswitch:PartType �0.54 0.85 �0.63 .530Repetition:DomLang:PartType �1.77 0.86 �2.06 .039Codeswitch:DomLang:PartType 1.49 0.88 1.70 .090Repetition:Codeswitch:DomLang:PartType �1.61 1.72 �0.94 .348



4c. Raw intensity (unbalanced bilinguals).


(Intercept) 61.54 0.77 80.35 <1e�03Repetition 0.21 0.13 1.66 .097Codeswitch 0.56 0.17 3.30 .001DomLang 0.23 0.15 1.49 .136PartType �0.18 1.52 �0.12 .908Repetition:Codeswitch �0.58 0.25 �2.30 .021Repetition:DomLang �0.06 0.25 �0.25 .805Codeswitch:DomLang 0.30 0.25 1.19 .236Repetition:PartType 0.16 0.25 0.65 .519Codeswitch:PartType 0.18 0.30 0.61 .543DomLang:PartType �0.84 0.52 �1.63 .104Repetition:Codeswitch:DomLang 0.21 0.50 0.42 .671Repetition:Codeswitch:PartType 0.43 0.50 0.85 .394Repetition:DomLang:PartType 0.71 0.50 1.42 .156Codeswitch:DomLang:PartType �0.49 0.59 �0.82 .410Repetition:Codeswitch:DomLang:PartType �1.44 1.01 �1.43 .153


4d. Relative intensity (unbalanced bilinguals).


(Intercept) 7.95 0.62 12.89 <1e�03Repetition 0.42 0.30 1.38 .167Codeswitch 0.94 0.34 2.77 .006DomLang 1.56 0.79 1.97 .049PartType 0.07 1.17 0.06 .954Repetition:Codeswitch �1.82 0.51 �3.58 <1e�03Repetition:DomLang 0.27 0.50 0.53 .594Codeswitch:DomLang 0.91 0.50 1.81 .070Repetition:PartType 0.68 0.60 1.13 .260Codeswitch:PartType �0.55 0.66 �0.83 .405DomLang:PartType �10.19 1.77 �5.76 <1e�03Repetition:Codeswitch:DomLang 0.81 1.01 0.80 .423Repetition:Codeswitch:PartType 0.19 1.01 0.19 .851Repetition:DomLang:PartType 0.98 1.01 0.97 .331Codeswitch:DomLang:PartType �1.52 1.05 �1.44 .150Repetition:Codeswitch:DomLang:PartType �1.42 2.03 �0.70 .483



References

Almor, A. (1999). Noun-phrase anaphora and focus: The informationalload hypothesis. Psychological Review, 106, 748–765.

Arnold, J. E. (1998). Reference form and discourse patterns. Unpublisheddoctoral dissertation, Stanford University.

Arnold, J. E., Kahn, J., & Pancani, G. (2012). Audience design affectsacoustic reduction via production facilitation. Psychological Bulletinand Review, 19, 505–512.

Aylett, M. P., & Turk, A. (2004). The smooth signal redundancy hypothesis:A functional explanation for relationships between redundancy,

prosodic prominence, and duration in spontaneous speech.Language and Speech, 47, 31–56.

Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effectsmodeling with crossed random effects for subjects and items. Journalof Memory and Language, 59, 390–412.

Bard, E. G., Anderson, A. H., Sotillo, C., Aylett, M. P., Doherty-Sneddon, G., &Newlands, A. (2000). Controlling the intelligibility of referringexpressions in dialogue. Journal of Memory and Language, 42, 1–22.http://dx.doi.org/10.1006/jmla.1999.2667.

Bard, E. G., & Aylett, M. (2004). Referential form, duration, and modelingthe listener in spoken dialogue. In J. Trueswell & M. Tanenhaus (Eds.),Approaches to studying world-situated language use: Bridging the

http://refhub.elsevier.com/S0749-596X(15)00073-X/h0005












http://dx.doi.org/10.1006/jmla.1999.2667





language-as-product and language-as-action traditions(pp. 173–191). Cambridge: MIT Press.

Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effectsstructure for confirmatory hypothesis testing: Keep it maximal.Journal of Memory and Language, 68, 255–278. http://dx.doi.org/10.1016/j.jml.2012.11.001.

Baumann, S., & Riester, A. (2013). Coreference, lexical givenness andprosody in German. Lingua, 136, 16–37.

Bell, A., Brenier, J. M., Gregory, M., Girand, C., & Jurafsky, D. (2009).Predictability effects on durations of content and function words inconversational English. Journal of Memory and Language, 60, 92–111.http://dx.doi.org/10.1016/j.jml.2008.06.003.

Boersma, P., & Weenink, D. (2007). Praat: Doing phonetics by computer(version 4.5.14) [computer program]. <http://www.praat.org/>.

Brainard, D. H. (1997). The psychophysics toolbox. Spatial Vision, 10,433–436.

Brennan, S. E., & Clark, H. H. (1996). Conceptual pacts and lexical choice inconversation. Journal of Experimental Psychology: Learning, Memory,and Cognition, 22, 1482–1496.

Clark, E. V. (1990). On the pragmatics of contrast. Journal of ChildLanguage, 17, 417–431.

Clark, E. V. (1997). Conceptual perspective and lexical choice inacquisition. Cognition, 64, 1–37. http://dx.doi.org/10.1016/S0010-0277(97)00010-3.

Costa, A., Caramazza, A., & Sebastian-Galles, N. (2000). The cognatefacilitation effect: Implications for models of lexical access. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 26,1283–1296.

Costa, A. (2005). Lexical access in bilingual production. In J. F. Kroll & A. M.B. De Groot (Eds.), Handbook of bilingualism: Psycholinguisticapproaches (pp. 308–325). New York: Oxford University Press.

Costa, A., & Santesteban, M. (2004). Lexical access in bilingual speechproduction: Evidence from language switching in highly proficientbilinguals and L2 learners. Journal of Memory and Language, 50,491–511. http://dx.doi.org/10.1016/j.jml.2004.02.002.

Dell, G. S. (1986). A spreading-activation theory of retrieval in sentenceproduction. Psychological Review, 93, 283.

Dell, G. S., & O’Seaghdha, P. G. (1991). Mediated and convergent lexicalpriming in language production. A comment on Levelt et al. (1991).Psychological Review, 98, 604–614.

Dell, G. S., & O’Seaghdha, P. G. (1992). Stages of lexical access in speechproduction. Cognition, 42, 287–314.

Fowler, C. A. (1988). Differential shortening of repeated content wordsproduced in various communicative contexts. Language and Speech,31, 307–319.

Fowler, C., & Housum, J. (1987). Talkers’ signaling of ‘‘new” listeners’perception and ‘‘old” words in speech and use of the distinction.Journal of Memory and Language, 26, 489–504.

Fraundorf, S. H., Diaz, M. I., Finley, J. R., Lewis, M. L., Tooley, K. M., Isaacs,A. M., et al. (2014). CogToolbox for MATLAB [computer software].<http://www.scottfraundorf.com/cogtoolbox.html>.

Galati, A., & Brennan, S. E. (2010). Attenuating information in spokencommunication: For the speaker, or for the addressee? Journal ofMemory and Language, 62, 35–51. http://dx.doi.org/10.1016/j.jml.2009.09.002.

Garrod, S., & Anderson, A. (1987). Saying what you mean in dialogue: Astudy in conceptual and semantic co-ordination. Cognition, 27,181–218.

Garrod, S., & Doherty, G. (1994). Conversation, co-ordination andconvention: An empirical investigation of how groups establishlinguistic conventions. Cognition, 53, 181–215.

Gollan, T. H., & Ferreira, V. S. (2009). Should I stay or should I switch? Acost–benefit analysis of voluntary language switching in young andaging bilinguals. Journal of Experimental Psychology: Learning, Memory,and Cognition, 35, 640–665.

Jun, S. A. (1993). The phonetics and phonology of Korean prosody.Unpublished Ph.D. dissertation, The Ohio State University,Columbus, Ohio.

Jun, S. A. (1998). The accentual phrase in the Korean prosodic hierarchy.Phonology, 15, 189–226.

Kahn, J. M., & Arnold, J. E. (2012). A processing-centered look at thecontribution of givenness to durational reduction. Journal of Memory

and Language, 67, 311–325. http://dx.doi.org/10.1016/j.jml.2012.07.002.

Kleiner, M., Brainard, D., & Pelli, D. (2007). What’s new in psychtoolbox-3? Perception, 36 (ECVP Abstract Supplement).

Kochanski, G., Grabe, E., Coleman, J., & Rosner, B. (2005). Loudnesspredicts prominence: Fundamental frequency lends little. The Journalof the Acoustical Society of America, 118, 1038–1054.

Kroll, J. F., Bobb, S. C., Misra, M., & Guo, T. (2008). Language selection inbilingual speech: Evidence for inhibitory processes. Acta Psychologica,128, 416–430. http://dx.doi.org/10.1016/j.actpsy.2008.02.001.

Kroll, J. F., & Stewart, E. (1994). Category interference in translation andpicture naming: Evidence for asymmetric connections betweenbilingual memory representations. Journal of Memory and Language,33, 149–174.

Lam, T. Q. (2012). The prominence of referring expressions: Message andlexical level effects. Unpublished doctoral dissertation, University ofIllinois at Urbana-Champaign.

Lam, T. Q., & Watson, D. G. (2010). Repetition is easy: Why repeatedreferents have reduced prominence. Memory & Cognition, 38,1137–1146. http://dx.doi.org/10.3758/MC.38.8.1137.

Lam, T. Q., & Watson, D. G. (2014). Repetition reduction: Lexical repetitionin the absence of referent repetition. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 40, 829–843.

Levelt, W. J. (1989). Speaking: From intention to articulation. Cambridge,MA: MIT Press.

Levelt, W. J., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access inspeech production. Behavioral and Brain Sciences, 22, 1–38.

Marian, V., Blumenfeld, H. K., & Kaushanskaya, M. (2007). The LanguageExperience and Proficiency Questionnaire (LEAP-Q): Assessinglanguage profiles in bilinguals and multilinguals. Journal of Speech,Language and Hearing Research, 50, 940.

Meagher, B. R., & Fowler, C. A. (2014). Embedded articulation: shifts inlocation influence speech production. Language, Cognition andNeuroscience, 29, 561–567.

Metzing, C., & Brennan, S. E. (2003). When conceptual pacts are broken:Partner specific effects on the comprehension of referringexpressions. Journal of Memory and Language, 49, 201–213.

Monsell, S., Matthews, G. H., & Miller, D. C. (1992). Repetition oflexicalization across languages: A further test of the locus ofpriming. The Quarterly Journal of Experimental Psychology Section A:Human Experimental Psychology, 44, 764–783. http://dx.doi.org/10.1080/14640749208401308.

Olson, D. R. (1970). Language and thought: Aspects of a cognitive theoryof semantics. Psychological Review, 77, 257–273.

Peterson, R. R., & Savoy, P. (1998). Lexical selection and phonologicalencoding during language production: Evidence for cascadedprocessing. Journal of Experimental Psychology: Learning, Memory,and Cognition, 24, 539.

Pluymaekers, M., Ernestus, M., & Baayen, R. H. (2005). Articulatoryplanning is continuous and sensitive to informational redundancy.Phonetica, 62, 146–159. http://dx.doi.org/10.1159/000090095.

Rapp, B., & Goldrick, M. (2000). Discreteness and interactivity in spokenword production. Psychological Review, 107, 460.

Rosa, E. C., Finch, K., Bergeson, M., & Arnold, J. E. (2015). The Effects ofAddressee Attention on Prosodic Prominence. Language, Cognition, andNeuroscience, 30, 48–56. http://dx.doi.org/10.1080/01690965.2013.772213.

Rossion, B., & Pourtois, G. (2001). Revisiting Snodgrass and Vanderwart’sobject database: Color and texture improve object recognition. Journalof Vision, 1, 413a.

Snodgrass, J. G., & Vanderwart, M. (1980). A standardized set of 260pictures: Norms for name agreement, image agreement, familiarity,and visual complexity. Journal of Experimental Psychology: HumanLearning & Memory, 6, 174–215. http://dx.doi.org/10.1037/0278-7393.6.2.174.

Terken, J., & Hirschberg, J. (1994). Deaccentuation of words representing‘‘given” information: Effects of persistence of grammatical functionand surface position. Language and Speech, 37, 125–145. http://dx.doi.org/10.1177/002383099403700202.

Watson, D. G., Arnold, J. E., & Tanenhaus, M. K. (2008). Tic Tac Toe: Effectsof predictability and importance on acoustic prominence in languageproduction. Cognition, 106, 1548–1557. http://dx.doi.org/10.1016/j.cognition.2007.06.009.








http://www.praat.org/








http://dx.doi.org/10.1016/S0010-0277(97)00010-3

http://dx.doi.org/10.1016/S0010-0277(97)00010-3
























http://www.scottfraundorf.com/cogtoolbox.html






















http://dx.doi.org/10.1016/j.actpsy.2008.02.001





http://dx.doi.org/10.3758/MC.38.8.1137


















http://dx.doi.org/10.1080/14640749208401308

http://dx.doi.org/10.1080/14640749208401308







http://dx.doi.org/10.1159/000090095



http://dx.doi.org/10.1080/01690965.2013.772213




http://dx.doi.org/10.1037/0278-7393.6.2.174

http://dx.doi.org/10.1037/0278-7393.6.2.174

http://dx.doi.org/10.1177/002383099403700202

http://dx.doi.org/10.1177/002383099403700202

http://dx.doi.org/10.1016/j.cognition.2007.06.009

http://dx.doi.org/10.1016/j.cognition.2007.06.009

Date post:	17-Aug-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

Journal of Memory and...

Documents